PoseNet - Split pose estimation methods  into estimateSinglePose and estimateMultiplePoses, and move inputResolution specification from loading to inference time. (#242)

PoseNet - to make a cleaner api, split the existing pose estimation method into estimateSinglePose and estimateMultiplePoses with an options object with parameters specific to that object.

Moved inputImageResolution into the inference params, since it's not needed at loaded time. In demos, no longer re-loading the model when inputImageResolution is changed.

Fixed documentation that had out of date input image resolution sizes.

Consolidated functionality in estimation methods into encaspulated functions that are portable and reusable.

Solidify typings of inference config objects.

Treating the config objects as immutable.

Fixed the posenet readme to have inference params being an object.

Also fixes the naming of maxDetections parameter in the readme.

Additionally, adds better typing to the inference config objects, and treats them as immutable when setting defaults.